Interesting, thanks! Well, as I said, I would expect them to outperform at large scales, specifically because they're learning and memoising large, stable patt

Interesting, thanks! Well, as I said, I would expect them to outperform at large scales, specifically because they're learning and memoising large, stable patterns (ed. in the sense of teleconnections) at low wavenumbers. I hope they have a switch to turn it off if we ever mess up and go back to a single-cell Hadley configuration :)